Inference and Attack in Bayesian Networks

In legal reasoning the Bayesian network approach has gained increasingly more attention over the last years due to the increase in scientific forensic evidence. It can however be questioned how meaningful a Bayesian network is in terms that are easily comprehensible by judges and lawyers. Argumentation models, which represent arguments and defeat, are arguably closer to their natural way of arguing and therefore potentially more easy to understand for lawyers and judges. The automated extraction of rules, arguments and counter-arguments from Bayesian networks will facilitate the communication between lawyers and judges on the one hand and forensic experts on the other. In this paper we propose a method to automatically extract inference rules and undercutters from Bayesian networks from which arguments can subsequently be constructed.